San Francisco (California), New York and London | Financial Times Nvidia is in talks to acquire US startup Reflection AI, a developer of open weight models that President Donald Trump's administration hopes will rival cheap Chinese alternatives, or expand its investment in the company, according to people with direct knowledge of the matter.
Talks are at an early stage, and a deal could take different forms, including a deal known as an "acqui-hire" in which Nvidia would hire employees and license technology without going through the protracted regulatory scrutiny associated with a full acquisition, the sources added.
Nvidia, which is already one of Reflection's largest shareholders, may also choose to strengthen its ties with the startup through a new capital injection or an agreement to provide it with more chips and computing capacity.
Advertising
A deal between the two companies could be reached in the coming weeks, according to several people familiar with the matter, although they have warned that negotiations could fail. It was not possible to determine the terms under discussion, but Reflection was last valued at US$25 billion (almost R$125 billion) in a financing round held in March.
Reflection and Nvidia declined to comment.
Advertising
"Acqui-hire" transactions have become increasingly popular among technology giants seeking to avoid antitrust scrutiny over business in the AI sector. Nvidia adopted a similar format with chip designer Groq, in a US$20 billion (R$100 billion) deal closed in December. This week, the business led to a lawsuit filed by Groq employees who felt aggrieved.
Nvidia, the most valuable company in the world, has been promoting a wave of investments in the last two years, in an attempt to strengthen the AI economy and consolidate its position at the center of this market. The company already has a family of open-weight AI models known as Nemotron, but does not have its own model at the technological frontier and on a large scale.
In recent months, the company has increasingly focused on fostering a market for open weight models, which users can customize and run on their own hardware. For businesses, they can be much cheaper to operate than more advanced versions of OpenAI's ChatGPT or Anthropic's Claude.
In July, Nvidia CEO Jensen Huang led the industry in advocating the creation of a national "ecosystem" of open weights, which he said would be essential to "creating opportunities for innovation and prosperity across the country." In September, the company valued at US$5.5 trillion (R$27.45 trillion) agreed to acquire Hugging Face, a repository of millions of models and datasets, for US$13 billion (R$65 billion).
Nvidia has pumped tens of billions of dollars into major AI labs, including OpenAI, "neocloud" companies like Nebius, and a host of smaller groups. This, in turn, has generated additional demand for its chips, which remain a crucial resource for companies developing or operating AI models.
C-Level
A newsletter with exclusive information for those who need to make decisions
Reflection is considered a strategically important asset for the United States in its AI race against Chinese rivals such as DeepSeek. This week, the company introduced its first public, customizable AI product, which runs on Nvidia chips. Reflection claims that the model is on the same level as the GLM-5.2, from Z.ai, one of the main Chinese competitors.
Advertising
The group's founders, Misha Laskin and Ioannis Antonoglou, previously worked on Gemini, Google's AI model.
Nvidia has already invested US$800 million (R$4 billion) in Reflection, which also has the support of venture capital companies such as Sequoia Capital, Disruptive and 1789 Capital, of which Donald Trump Jr., Trump's son, is a partner.
